Module canonical 

Canonical simplification: recognizing that exact geometry is secretly analytic: a B-spline surface that is a cylinder becomes the cylinder.

The input is exact geometry, not samples of unknown provenance: the decision is made against the surface’s own equation, sampled on its own chart, with its own normals. That is what separates this from reverse engineering: nothing is guessed about what the data means, only checked against a candidate the estimators propose. A candidate is accepted when every sample sits within the caller’s stated tolerance of it, and the certificate is the worst deviation actually measured; a surface that is genuinely free-form at that tolerance stays what it is.

Why it matters: exchange formats routinely spell a cylinder out pointwise, and every algorithm downstream (intersection, blending, the boolean’s closed forms) is faster and exacter on the analytic carrier. The reference keeps this in its healing layer for the same reason.

The estimators are classical, and shared with the sample-based recognizer that lives outside the kernel: a plane is the mean normal; a sphere’s centre is where the normal lines meet, in least squares; a cylinder’s axis is the direction the normals avoid (their covariance’s smallest eigenvector), and a cone adds the linear taper of radius against height.

Structs

CanonicalReport

A report of what was simplified, face by face.

Enums

Simplified

What one face’s simplification found.

Functions

canonical_simplify

Replace every free-form surface in shape that is secretly analytic, within tolerance, by the plane, cylinder, cone or sphere it is.

recognize_surface

The candidate the estimators propose, verified against every sample.